基于小波时频图与卷积神经网络的高层建筑结构损伤定位OA

Story-level damage localization for high-rise building structures is of great significance for ensuring their operational safety.Therefore,a multi-label classification damage localization method based on wavelet time-frequency diagrams and convolutional neural networks was proposed.Firstly,a dual attention mechanism was introduced based on the VGG16 architecture to construct a lightweight improved model VBAG-Net,which enhanced the learning ability for sensitive features of damage.Secondly,damage experiments for a semi-prefabricated steel frame-core tube model was designed and completed where structural damage was simulated by disassembling wall components,establishing a benchmark database for story-level damage identification in high-rise structures.The one-dimensional vibration signals were converted into two-dimensional time-frequency diagrams using the continuous wavelet transform(CWT)and synchrosqueezed wavelet transform(SWT)as the model input,respectively.Finally,the VBAG-Net was trained and tested based on this dataset,and compared with several mainstream CNN models.The results show that VBAG-Net achieved exact match ratio of 98.48%(CWT)and 98.86%(SWT)on the test set.It still maintains excellent performance even when varying the time-segment length of vibration signals for time-frequency diagram generation.The comprehensive identification effect and robustness are significantly superior to the comparative models,verifying the effectiveness of this method in locating story-level damage in high-rise buildings.